GEOlimma: differential expression analysis and feature selection using pre-existing microarray data

Area of Science:

• Transcriptomics
• Bioinformatics
• Computational Biology

Background:

• Differential expression and feature selection are crucial for disease classifiers using transcriptomics data.
• Challenges include high dimensionality and data noise.
• Incorporating pre-existing transcriptomics data can improve gene identification and classification.

Purpose of the Study:

• To develop a novel method, GEOlimma, for differential gene expression and feature selection.
• To leverage existing Gene Expression Omnibus (GEO) data with the Limma method.
• To enhance the identification of differentially expressed (DE) genes by utilizing experimental condition information.

Main Methods:

• GEOlimma combines pre-existing microarray data from GEO with the Limma method.
• Quantified differential gene expression across 2481 pairwise comparisons from 602 GEO datasets.
• Converted differential expression frequencies to DE prior probabilities.

Main Results:

• GEOlimma identified DE genes enriched in cell growth, death, signal transduction, and cancer pathways.
• Applied GEOlimma to human disease datasets, showing increased experimental power compared to Limma.
• Achieved similar or better classification performance than Limma on an asthma dataset using GEOlimma for feature selection.

Conclusions:

• GEOlimma is a more effective method for differential gene expression and feature selection than standard Limma.
• GEOlimma enhances experimental power and classification performance.
• The method has potential applications beyond transcriptomics to other high-throughput biological datasets.